Effect of Bit Depth on Cloud Segmentation of Remote-Sensing Images

Due to the cloud coverage of remote-sensing images, ground object information will be attenuated or even lost, and texture spectral image changed at same time. Accurately detecting clouds from images is great significance field remote sensing. Cloud detection utilizes semantic segmentation classify pixel level. However, previous studies have focused on improvement algorithm performance, little attention has been paid impact bit depth detection. In this paper, deep UNet taken as an example, a set widely used labeling dataset “L8 Biome” verification data explore relationship between accuracy different surface landscapes when for The research results show that normalized, effect with 16-bit slightly better than 8-bit sensing image; not gap widened. using training take longer. This means selection classification do always need follow highest possible doing but should consider balance efficiency accuracy.